Jeff Fessler

Thanks for the very quick reply! And thanks for the package too - I was not familiar with the Kahan method prior to seeing it used here. Related functions like...

Sorry I thought I had included a link in an earlier reply. Here is a detailed Literate example that uses `fit(Histogram,...)` and that specifically refers to this issue: https://juliaimagerecon.github.io/Examples/generated/mri/1-nufft/ There...

I gave a MWE in the OP. If you have any examples of use cases where one needs `Real` instead of `Int` weights then I am sure that in those...

Sorry, I just realized that I reported the error in the OP, but I did not describe well enough the desired behavior. Here is a MWE of approximately how I'd...

> Unless we add a special method for complex numbers, there's no way to decide what to do with the real and imaginary parts in order to choose edges which...

> the unit should go with data right? Good point, but supporting units for the data is another issue for later. This issue is about relaxing the type of the...

> I've never seen non-real weights, and I personally can't imagine I hope that the packages I am developing in Julia will be used in ways I have never seen...

Let me try to elaborate here. Once we go from `Int` to `Real`, we are going beyond *counting* to *binned accumulation*. Binned accumulation is useful beyond statistics, including in the...

> So unless there's a conceptually sound / existing domain-specific application, I don't think weight can be non-real number. Above I have 1) provided pointers to a Julia code repo...

A PhD student in my group adapted the gist code provided by @sdewaele into this package where we use ForwardDiff to autodifferentiate matrix exponentials: https://github.com/StevenWhitaker/BlochSim.jl/blob/main/src/expm.jl However, the original code was...